AI systems are uncertain about their outputs more often than their confident-sounding responses suggest. A language model that states something as fact may be drawing on sparse or contradictory training data. A classification model that labels something as "positive" might have assigned it a 52% probability - barely better than a coin flip. Communicating this uncertainty to users is both critically important and surprisingly difficult. Most people don't naturally think in probabilities, and presenting raw confidence scores can confuse rather than inform. Saying "the model is 73% confident" sounds precise but doesn't help someone decide whether to act on the recommendation. Effective uncertainty communication uses multiple strategies: visual indicators like confidence bars, verbal hedging ("the model suggests, but is less certain than usual"), highlighting cases where the system's input data was limited, and explicitly flagging when a query falls outside the system's areas of strength. The goal isn't to make people distrust every output, but to give them the information they need to apply the right level of scrutiny to each one.